Overview
We have worked to describe the social patch structure of the Baltimore Metropolitan Region with several goals in mind. Social patch structure GIS databases are constructed to link to spatially-explicit social processes and biophysical structures and processes. Social patch structure GIS databases are time-series so that 1) changes in social patch structure can be measured over time, 2) cause and effect relationships with social processes and biophysical structures and processes can be assessed, and 3) temporal complexity in terms of lags, legacies, slow processes, and system resiliency and stability can be elucidated.
We have adopted two strategies for our time-series, social patch structure GIS databases. We use pre-defined social patches such as counties and neighborhoods (US Census Block Groups) and existing attribute data and 2) empirically derived social-ecological patches, based upon varying relationships among variables over space and time.
To date, a significant portion of our effort has focused on the acquisition and organization of existing, long term, spatially-explicit demographic and socioeconomic data. From a variety of sources such as Sanborn Fire Insurance Maps, business atlases, state forestry maps (1914), and aerial imagery (1938, 1957), we have developed land use maps from 1876 to the present. From the US Census, we have data such as total population and population demographics, total households, household composition, educational attainment, household income, race/ancestry, occupation, employment, housing type, housing value, homeownership, and household residence time. We have developed these data at the County level for the entire Baltimore Metropolitan Region from 1790 to 2000, at the US Census Tract level for Baltimore City from 1940 to 2000, and at the US Census Block Group level from 1970 to the Present.
We have also used Claritas' PRIZM (Potential Rating Index for Zipcode Markets) classification system to characterize neighborhoods at the US Census Block Group level. PRIZM has been developed by demographers and sociologists for market research (Weiss 1988; Grove and Burch 1997; Weiss 2000). There are two primary goals of the original PRIZM classification system. The first goal is to categorize the 250 million people of the American population and their urban, suburban, and rural neighborhoods into lifestyle clusters using Census data about household education, income, occupation, race/ancestry, family composition, and housing. The second goal is to associate these clusters with characteristic household tastes and attitudes using market research surveys, public opinion polls, and point-of-purchase receipts. For instance, Mediamark Research, Inc. and Simmons Market Research Bureau collect information about household lifestyle preferences and radio and TV programs. The Polk Company compiles car and truck statistics from motor vehicle registrations. And NFO Research, Inc. conducts surveys that are used to track political and social issues (Weiss 1988; Weiss 2000).
In the PRIZM system, each cluster is a statistical entity, synonymous with a "class" in any standard classification system. The classification system can be disaggregated from 5 to 15 and 62 classes. The five classes are arrayed along an axis of urbanization. From 5 to 15 classes adds a second axis: socioeconomic status. The 62 class classification further expands the socioeconomic status axis with components including household composition, mobility, ethnicity, and housing characteristics (Claritas 1999). In addition to its utility for characterizing what people are likely to prefer, PRIZM is widely generalizable since this classification system has been applied and evaluated on a national and global basis (Weiss 2000).
The PRIZM classification system in its original form is extremely useful for our efforts for a number of reasons. First, PRIZM essentially represents a spatially-explicit classification of group identity and social status based upon reference group behavior theory (Merton and Kitt 1950; Shibutani 1955; Hyman and Singer 1968; Singer 1981) and consumer behavior data in terms of household preferences for a wide spectrum of market and non-market goods and services (Grove and Burch 2002). Second, because PRIZM is designed to predict household preferences for market goods and services, it is well-suited for understanding variations in household land-management preferences and behavior using lawncare purchasing data from companies such as Mediamark, Inc.. Third, PRIZM is useful at the neighborhood level because every US CBG is assigned a specific PRIZM class value. Finally, by converting the necessary historic Census geographies and attribute data for Baltimore from the 1880s to the present, a historical "pseudo-PRIZM" approach is being developed for long-term analysis of urban neighborhood change.
Our adaptation and application of what we call PRIZM-SE (social ecology) extends beyond interpretations of existing, marketing data and includes enhancements through the collection and incorporation of supplemental, social and ecological primary data. These supplemental data are necessary because PRIZM does not capture all of a neighborhood's social and ecological qualities. As Weiss (2000:181) notes:
…there are many characteristics missed in this portrait-by-numbers. The clusters reveal little about the friendliness of an area, whether trees provide shade when someone walks a dog, if the children are happy. And the portraits may not exactly match every household in a neighborhood because the cluster system operates on the law of averages, providing tendencies and correlations for a group of households. What the cluster profiles do reveal is how people behave in the public realm: what they buy in stores, where they play after work, how they vote at the ballot box, and -- more importantly -- how they compare to others from different lifestyles." (italics added)
The items Weiss identifies as missing are exactly some of the social and ecological processes in which we are interested. For instance, are there differences among PRIZM clusters in terms of environmental quality and management: air, water, safety, lawns, trees, and gardens? Answering these questions is crucial for understanding how social groups differentially affect ecological structure and function.
In addition to existing social patch delineations and characterizations, we are developing a novel method to derive social patches from behavioral data using Geographically Weighted Regressions (GWR). For example, one of the most challenging problems in housing economics has been the delineation of housing submarket boundaries (Dale-Johnson 1982; Bourassa, Hoesli et al. 2003; Goodman and Thibodeau 2003; Thibodeau 2003). This is because housing markets are not directly observable. They are dynamic entities defined by a combination of geography and socio-economic processes or relationships. Higher order statistical measures are needed to map these relationships over space. Using the results of GWR hedonic regressions, parameter estimates and test statistics can be plotted over space to delineate areas of relative homogeneity. For instance, plotting the coefficient on some location features, like surrounding tree cover, would show the variability in willingness to pay (WTP) for tree cover over space. Values for non-sampled locations can be interpolated using geostatistics. Once parameters/test statistics have been encoded for each pixel, cluster analysis is used to delineate areas with similar parameter combinations.
